Developers introduced self-hosted API jeff based on GLiFormer

Developers introduced jeff — a system for implementing TypeSafe System One API, focused on local deployment. The tool is based on the GLiFormer architecture with 400 million parameters and provides support for selection, evaluation, and noul operations.
The system is designed to work in isolated environments and supports deployment through the Modal platform or local launch. To function, the project requires Python 3.12 and the uv package manager. During testing, the solution demonstrated a result of 8 points when performing tasks in 3 hours.
The API implementation allows developers to integrate typed operations into their own AI pipelines without dependence on external cloud providers. The use of GLiFormer ensures a balance between computational complexity and accuracy of request processing within the given operations.
The project is focused on creating predictable interaction interfaces with models, where strict data typing is a critical requirement. The local nature of the implementation eliminates delays in data transmission and allows controlling the entire request processing cycle within the user's infrastructure.



